Short answer: trust AI study answers only as far as you can check them. Tools like ChatGPT can state wrong facts fluently, especially numbers, names, dates, definitions and references. Before you memorise anything, check it against your own lecture notes or textbook: the answer should be backed by a specific sentence in your material. If the AI can't point to where your notes say it, treat the answer as unconfirmed.
Why AI gives confident wrong answers
A language model like ChatGPT writes the most likely next words, based on patterns in the text it learned from. Most of the time that produces correct answers. But it has no built-in sense of which of its sentences are true, so when it doesn't know something it can still produce a fluent, plausible answer. This is often called a hallucination.
It is measurable. When Walters and Wilder (2023) asked ChatGPT to write short literature reviews with references, 55% of the citations produced by GPT-3.5 and 18% of those produced by GPT-4 did not exist. Newer models are better, but the lesson holds: fluency is not evidence of accuracy.
Students feel the same worry. In a survey of UK undergraduates, getting false or biased results was one of the main reasons students gave for being put off using AI (HEPI, 2025).
Which answers to double-check
- Numbers and units: values, constants, percentages, doses of anything, dates.
- Names and terms: who discovered what, exact technical terms, spellings.
- Definitions your course defines its own way. Your lecturer's definition is the one that gets marked.
- Lists and steps: missing or extra items, wrong order.
- References and quotes: check that the source exists and says what is claimed.
- Anything your notes don't mention. It may be correct, but it isn't what you'll be examined on.
A 60-second way to check
- Ask for the source line. "Quote the exact sentence from my notes that supports this answer."
- Find it yourself. Search your notes (Ctrl+F or Cmd+F) for a few words of that quote. If you can't find it, the answer is unconfirmed.
- Read the sentence around it. Make sure the quote actually says what the AI claims, not something close to it.
- Mark anything unconfirmed. Ask your lecturer, check the textbook, or leave it out of your revision.
This takes under a minute per answer and catches most problems. It is also good revision in itself, because finding the line means going back into your material.
Why a wrong answer is worse than none
Testing yourself is one of the best ways to learn, but it works on whatever you practise, including mistakes. Research on multiple-choice tests found that students sometimes later reproduced the wrong options they had seen as if they were true, an effect that shrank when they got feedback with the correct answer (Roediger and Marsh, 2005). If your practice questions contain errors and nothing corrects them, you can learn the error well. That is why checking matters more for study material than for almost anything else you use AI for.
What to look for in an AI study tool
- It works from your material, not general knowledge.
- It shows you where each answer comes from, so you can check it in seconds.
- It rejects what it can't back up, rather than guessing.
- It gives feedback after every answer, so mistakes get corrected rather than learned.
That is the rule Ace is built around: it writes questions from your own PDF, slides or lecture video, checks that each answer is backed by a sentence copied word for word from your file, drops any question that fails, and shows you that sentence under every answer. More on that in how Ace writes questions.
Frequently asked questions
Is ChatGPT accurate enough to study with?
It is often right, but not reliably enough to memorise without checking. Use it to generate practice questions from your own notes, and verify answers against those notes.
How do I know if ChatGPT made something up?
Ask it to quote the sentence from your material that supports the answer, then find that sentence yourself. If it isn't there, treat the answer as unconfirmed.
Are AI flashcards accurate?
It depends on the tool. Look for one that builds cards only from your material and shows the source for each answer, so you can check it.
Should I stop using AI for studying?
No. Use it to quiz you on your own material, check what matters, and keep the thinking for yourself.
Sources
- Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13, 14045.
- Freeman, J. (2025). Student Generative AI Survey 2025. Higher Education Policy Institute (HEPI) and Kortext.
- Roediger, H. L., & Marsh, E. J. (2005). The positive and negative consequences of multiple-choice testing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(5), 1155–1159.



